AI didn't save any time. It just decided who spends it.
At least, in some areas of the business...
AI is not improving internal efficiency... it’s redistributing the cost of it.
This is not a complaint about the tools. AI is genuinely useful for research, for stress-testing ideas, for turning a rough brief into a structured draft. The problem is specifically what happens after the draft exists. Someone prompts an AI for twenty minutes and produces a fifteen-page strategy paper. They share it “for review and comment.” The twenty minutes they spent has just become four hours of reading time distributed across every person they CC’d. The work didn’t disappear... it transferred, invisibly, from the author to the audience.
This is not an isolated observation. The media is full of it right now. A New Zealand tenant recently used AI to build a $40,000 tenancy tribunal claim that ran to 215 pages, two hearings, and months of process... the tribunal awarded her $80. HBR and Stanford researchers have named the broader workplace pattern “workslop,” estimating the cost to large organizations at over $9 million a year in lost productivity, with 53% of recipients reporting annoyance and roughly half viewing the sender as less capable and reliable as a result. UC Berkeley found that AI doesn’t reduce work at all... it intensifies it, with employees working faster, across a broader scope, for longer hours, without anyone asking them to.
The to-do list problem is a specific version of the same thing. AI makes it trivially easy to generate exhaustive task lists... forty, sixty, a hundred items, fully detailed, logically structured, professionally formatted. Handing that list to a team feels like serious planning, except the hard work in planning is not listing tasks. It’s working out what a real team with a real workload can actually deliver, making the trade-offs, and standing behind the prioritization. AI can produce the list in thirty seconds. It cannot do that thinking. And when leaders skip that thinking, they’ve handed their teams an execution problem disguised as a plan.
The technical paper problem is worse. Non-technical people can now generate extensive whitepapers, architecture documents, and technical reference material in minutes, and then hand them to engineering with “I created this, please review and correct as needed.” That framing makes it sound like a small ask. It isn’t. Every single technical claim in an AI-generated document has to be validated by someone with the actual expertise to know whether it’s accurate, because AI hallucinations in technical content are still common even with the best models available. The author’s credibility cost is zero... the validation cost falls entirely on the people who know enough to catch the errors. That’s not a productivity gain for the organization. It’s a productivity transfer with a smile on it.
Brevity used to be a signal. A tightly written email, a one-page proposal, a three-sentence ask... these communicated something beyond the content itself. They told you the author had done the work to understand what actually mattered. Distillation was a skill, and it was respected because it was hard. AI has automated the appearance of that skill without delivering the substance of it, and the result is organizations drowning in content that looks comprehensive but carries no real intellectual accountability.
If you produce a document, you own it. Not just the output, but the research behind it, the reasoning, the trade-offs, the implications. When your team asks questions, you need to be able to answer them live. If you can’t, the document should not have been sent. Five hundred words is enough for almost any internal proposal. If you need more than that to make your case, the case probably isn’t ready yet... and no amount of AI-generated padding will change that.
The same dynamic plays out in engineering teams, where non-technical staff are now prompting their way to working applications and handing them to engineering as finished deliverables. The app works on Tuesday afternoon. The security vulnerabilities, dependency updates, and maintenance obligations that come with it last for the life of the business... but that's another discussion, and one I'll cover separately.
#rantover